CASANOVA: Permutation inference in factorial survival designs

Area of Science:

• Biostatistics
• Survival Analysis
• Clinical Trial Design

Background:

• Factorial designs are crucial for studying multiple interventions simultaneously.
• Time-to-event endpoints are common in clinical research, but analysis methods can be restrictive.
• Existing models often assume proportional hazards, limiting their applicability when curves cross.

Purpose of the Study:

• To develop flexible inference procedures for factorial designs with time-to-event outcomes.
• To enable the detection of crossing survival or hazard curves without power loss.
• To provide a distribution-free method for robust analysis.

Main Methods:

• Formulating null hypotheses based on cumulative hazards.
• Utilizing quadratic forms of Nelson-Aalen-type integrals to measure deviations.
• Employing a permutation strategy for distribution-free inference.
• Proving asymptotic validity of the proposed procedures.

Main Results:

• The proposed methods do not require restrictive assumptions like proportional hazards.
• Crossing survival or hazard curves can be detected with maintained statistical power.
• Extensive simulations demonstrate good performance in small samples.
• The approach is illustrated with an asthma dataset.

Conclusions:

• The developed inference procedures offer a powerful and flexible alternative for analyzing time-to-event data in factorial designs.
• The method enhances the ability to detect complex survival patterns, such as crossing curves.
• The approach is statistically valid and practically applicable in various research settings.